Saccadic accuracy, measured as the ratio of the size of a saccade to the size of the target step that elicits it, i.e., saccade gain, can be altered by jumping the target surreptitiously during the targeting saccade. The gain change produced by this paradigm does not generalize or transfer to saccades of all sizes. Instead, the amount of transfer decreases the more the tested saccade differs in amplitude and direction from that adapted. Here, we tested the limits of this saccade-size specificity by attempting to impose quite different gain states on saccades in the same direction. We altered the saccadic gain by intrasaccadic target jumps of 30% of the initial target step, either forward to produce a gain increase or backward to produce a gain decrease. Three different conditions were studied: (1) saccades to target steps of 20 degrees or 7 degrees were adapted in individual sessions with backward and forward jumps, respectively; (2) saccades to target steps of 20 degrees caused backward target jumps during the same session in which saccades to 7 degrees target steps caused forward steps; (3) the target jumps accompanying 20 and 7 degrees saccades were the same as in (2), but in addition, there were intermediate-sized saccades to 13.5 degrees target steps with no intrasaccadic target jumps. Saccadic gain adaptation was quite flexible. In condition 2, we could simultaneously increase the gain of saccades to 7 degrees target steps while decreasing the gain of saccades to 20 degrees steps in the same direction. Intermediate horizontal saccades to 13.5 degrees target steps experienced gain reductions (average: 6.9%), which were not the sum of gain changes expected from separate 20 degrees gain decreases and 7 degrees gain increases alone, as predicted from condition 1. If adaptation at 20 degrees and 7 degrees occurred while an animal also tracked a non-adapting 13.5 degrees target step (paradigm 3), the gain reduction of saccades to the 13.5 degrees step was reduced considerably (3.4%). Thus, the mechanism that adapts saccade size can support a robust gain increase for saccades of one size while simultaneously supporting a robust gain decrease for saccades only 13 degrees larger. Furthermore, the presence during adaptation of a non-adapted target step with a size intermediate to the two adapting steps reestablishes a nearly normal gain within only 6.5 degrees of a robust gain increase and decrease. These data indicate that saccadic gain adaptation can set very different gain states for saccades with rather similar vectors.